Hangzhou, China – September 22, 2026
Alibaba Group’s T-Head semiconductor unit is set to launch its most powerful domestic AI chip yet, the Zhenwu V900, at the ongoing 2026 Yunqi Conference in Hangzhou. This significant announcement marks a substantial leap in China’s indigenous AI chip capabilities, promising a threefold increase in computing power compared to its predecessor, the Zhenwu M890.
Zhenwu V900: A Leap in AI Performance
The Zhenwu V900 is engineered to deliver enhanced AI performance, boasting a computational power three times that of the previous Zhenwu M890. This advancement is crucial for the demanding tasks of training and inferencing cutting-edge AI models. Furthermore, the architecture of the Zhenwu V900 allows for the construction of single AI clusters scalable to an impressive 500,000 cards, enabling unprecedented scale for AI development and deployment.
T-Head’s Comprehensive Data Center Chip Strategy
The development of the Zhenwu V900 is part of T-Head’s broader strategy to establish a complete ecosystem of data center chips. The company has already made significant inroads with its existing product lines, which include the Zhenwu series GPUs, Yitian series CPUs, and Panmai series intelligent network interface cards, alongside ICN interconnect chips. This comprehensive portfolio aims to cover all core chip requirements for building ultra-large-scale AI clusters.
Alibaba anticipates a substantial increase in T-Head chip shipments, driven by the maturity of its chip product lines and their widespread adoption by customers. This growth is a testament to the company’s investment and progress in the highly competitive AI hardware market.
Synergy and Optimization for AI Clusters
Beyond the hardware itself, Alibaba is focusing on the synergistic optimization of its AI infrastructure. The company is actively fine-tuning the integration of its chips, servers, ultra-nodes, networking components, models, and inference systems. This holistic approach is designed to maximize the Token throughput and overall efficiency of its AI clusters.
A key component of this optimization is the self-developed M890 AI ultra-node, which has already demonstrated its capability in efficiently handling the inference of large models with over two trillion parameters. These ultra-nodes are being made available at scale on Alibaba Cloud, providing developers and businesses with powerful tools for their AI endeavors.
Expanding Cloud Services and Future Outlook
In a move to further support the growing demand for AI computing power, Alibaba Cloud plans to add new service nodes in the fourth quarter of this year. This expansion will increase the supply capacity of its ultra-nodes, ensuring that more users can leverage Alibaba’s advanced AI infrastructure.
The launch of the Zhenwu V900 and the ongoing optimizations across Alibaba’s AI ecosystem highlight the company’s commitment to pushing the boundaries of artificial intelligence and solidifying its position as a leader in both AI hardware and cloud services.
